Google is finally answering one of the most common complaints about Gemini: the visible watermark stamped on AI-generated images, videos, and music. A new setting called Media Watermarks gives users control over whether that visible tag appears on their creations. The update applies to content made with Nano Banana, Omni, and Lyria, and it is rolling out over the next few days.
Until now, the watermark was automatic and unavoidable. Anyone generating an image with Nano Banana or a video with Omni had to accept the visible marker. The new toggle changes that by letting users decide for themselves. Going forward, whatever choice is made in the settings will apply to every future image, video, and music track generated through the app.
How to turn off AI watermarks in Gemini
The process is straightforward. Open Gemini's settings, find the Media Watermark option, and flip the toggle. Turning it off removes the visible watermark from generated content. Turning it on restores it. The setting is global, meaning there are no per-file or per-project controls. Once the preference is saved, it stays in effect for all subsequent creations.
This may sound like a simple quality-of-life change, but it marks a significant shift in how Google handles provenance. Even with the visible tag disabled, the company continues to embed two layers of identification. Invisible SynthID watermarks are baked into the pixels or audio data, and C2PA metadata follows the file wherever it travels. That means the generated content remains traceable behind the scenes, even to someone who never notices the missing visible watermark.
What still stays embedded
SynthID is Google DeepMind's watermarking system. It works by embedding a pattern into the generated content that is imperceptible to the human eye but detectable by Google's scanner. Unlike a visible stamp, SynthID cannot be easily cropped out or overlaid. It is designed to survive compression, filtering, and many other transformations. C2PA, on the other hand, is the Coalition for Content Provenance and Authenticity standard. It adds metadata to the file that records when and how the content was created, as well as what tool produced it. This metadata can be viewed by anyone with a compatible tool, making it a useful way to verify whether an image or video originated from an AI system.
The combination of SynthID and C2PA means that switching off the visible watermark does not make the content anonymous or untraceable. Google has been clear about this from the beginning. The company has repeatedly emphasized that transparency around AI-generated content is key to fighting misinformation and building trust. By removing only the visible marker, Google is trying to satisfy users who find the stamp intrusive while still preserving a layer of accountability.
Availability and restrictions
The rollout is not universal. Google says the Media Watermark setting will become available to users worldwide over the next few days, but there are exceptions. In India, South Korea, and Vietnam, the option to disable visible watermarks is limited to subscribers of the paid AI Ultra plan. Free users in those countries will continue to see the watermark on generated content. There is no word on whether this restriction is permanent or tied to regulatory requirements in those markets.
Work and school accounts face a harder restriction. The toggle does not appear at all for these accounts, regardless of whether they have a paid plan. That means visible watermarks remain mandatory for users on Google Workspace and education accounts. Google has not explained the reasoning, but it may be related to enterprise content policies or legal obligations in the academic and corporate sectors.
This uneven rollout could lead to confusion. A user in one country might share a watermark-free image with a user in another country, and the recipient could assume the image was human-made. Google's invisible watermarking is meant to address that, but not every platform checks for SynthID. C2PA metadata, however, is increasingly supported by major image databases and social platforms. As adoption grows, the metadata becomes a more reliable way to confirm AI provenance.
Why Google made this change
The update is a direct response to user feedback. Google VP Josh Woodward previously asked people on social media to share their biggest frustrations with Gemini. Thousands of responses poured in, and removing the visible watermark from Nano Banana made the top 10 list. Woodward acknowledged the request and said the team would consider it. The arrival of the Media Watermark toggle is the result of that process.
It is worth noting that community efforts to remove watermarks from AI-generated content have been growing. Tools and scripts that automatically strip visible watermarks from images have circulated online, sometimes in direct violation of terms of service. By introducing an official toggle, Google gives users a legitimate way to remove the watermark without resorting to third-party hacks. This may reduce the appeal of those unofficial tools and bring the behavior back inside Google's control.
A closer look at AI watermarking trends
The move places Google alongside other major AI companies that are experimenting with provenance tools. OpenAI, for example, has been developing its own watermarking methods for text and imagery. Anthropic has also adopted C2PA-style metadata in some products. The general direction of the industry is toward robust provenance rather than visible, annoying stamps. Visible watermarks can ruin the aesthetic of an image or video, which might discourage creators from using a tool. Invisible watermarks preserve the creative experience while still offering a safety net.
There is also a broader regulatory angle. Governments around the world are beginning to draft rules around AI-generated content. The European Union's AI Act, for instance, includes transparency obligations that require AI providers to label synthetic content. While the exact technical standard varies, the combination of metadata and watermarks is likely to become the norm. Google's dual-layer approach positions it well for these legal requirements. Even if a visible watermark is turned off, the presence of SynthID and C2PA satisfies most transparency mandates.
Still, there are concerns. Some critics argue that invisible watermarks are not enough to prevent malicious use. A bad actor could edit an image, strip metadata, and resample it to weaken SynthID. Google acknowledges that no watermarking system is completely foolproof. The company describes SynthID as robust but not impenetrable. The goal, according to DeepMind, is to raise the cost of tampering and make it easier for platform moderators and fact-checkers to identify AI content when they need to.
Another angle is user education. Many people who use AI image generators may not understand the difference between visible and invisible watermarks. A user who disables the visible marker might assume that the content is now indistinguishable from human-made media. That assumption could cause problems if the content is used in journalism, advertising, or political campaigns. Google's user interface currently does not appear to display a prominent warning when the visible watermark is turned off, though the company has clearly explained the limitation in its announcement. In the future, we may see more explicit notifications or even conditional disabling based on the content's intended use.
The addition of the Media Watermark toggle is a small change in the interface but a meaningful one for creators. It gives them more control over their work, a common theme in the evolution of creative tools. For years, photographers have had to balance the desire for visibility with the annoyance of watermarks. AI creators face the same tension. Google is effectively saying that creators can remove the stamp, but the origin of the work is still recorded. This compromise is likely to be appreciated by both the casual user and the professional designer, even though the availability will vary by account type and region.
As the rollout continues, users should check their settings to see whether the option has arrived. The timing of the update is tied to Google's broader push to make Gemini a more consumer-friendly assistant. Removing a friction point that generations of users complained about sends a clear message: Google is listening. Whether that is enough to keep users from seeking out third-party tools remains to be seen, but the official toggle is a step in the right direction. With invisible provenance still in place, Google can argue that it has not sacrificed accountability for convenience. And with the visible watermark now optional, creators can present their AI-generated work without an ugly stamp, while Google still maintains a behind-the-scenes record of every piece of content created in Gemini.
Source: Digital Trends News